[jira] [Commented] (SAMZA-252) Document stream reprocessing

["Martin Kleppmann (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SAMZA-252?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13987106#comment-13987106 ] 

Martin Kleppmann commented on SAMZA-252:
----------------------------------------

So far it doesn't talk about the details of implementing a two-job system (where one job processes real-time updates and the other reprocesses history). We could explain the "A/B streams" approach that was discussed offline, and how a downstream job can use timestamps of jobs and/or version numbers of jobs to ensure that old data doesn't overwrite new data.

It would also be nice not to have to mess with checkpoints, and instead have a simple config for resetting to offset 0 when a new version of a job is deployed (e.g. by including the job version number/UUID in the checkpoint, and ignoring the checkpoint on mismatch). However, we don't have that implemented yet, so I didn't want to mention it in the docs.

If we include everything, this is going to get quite long — which is maybe a good thing, but it becomes a lot of work for people to read. Anyone have thoughts on where we should strike the balance between comprehensiveness and conciseness in this doc?

> Document stream reprocessing
> ----------------------------
>
>                 Key: SAMZA-252
>                 URL: https://issues.apache.org/jira/browse/SAMZA-252
>             Project: Samza
>          Issue Type: Improvement
>          Components: docs
>    Affects Versions: 0.6.0
>            Reporter: Chris Riccomini
>            Assignee: Martin Kleppmann
>             Fix For: 0.7.0, 0.8.0
>
>         Attachments: SAMZA-252.1.patch
>
>
> A need with stream processing is to want to re-process prior messages at some later date. An example of this is having a stream processing job that is classifying messages in some way using a machine learning algorithm. At some point, the algorithm will be updated with a more accurate vector of weights. When this happens, usually you wish to re-process past messages to get more accurate results. Usually this is solved by running a parallel pipeline from Hadoop.
> We have thought extensively about this use case, and should document how to use Samza in a re-processing use case.



--
This message was sent by Atlassian JIRA
(v6.2#6252)